The present invention relates to a video tape recorder for recording high definition television (HDTV) signals and in particular to a recorder which records motion compensated compressed high definition television signals and yet allows trick-play features such as viewing during fast-forward and reverse.
Digital high definition television production schemes have been under development for several years. These schemes typically produce higher quality television images and sound than conventional television systems. To do this, these systems convey more information than is conveyed in a conventional television signal.
Recent regulations issued by the Federal Communications Commission (FCC), however, require that this larger amount of information be sent within the same band of frequencies as a conventional television signal. To fit the relatively large band width High Definition Television (HDTV) signal into a standard television channels the information content of the HDTV signal must be compressed.
Because of the relatively large degree of spatial redundancy and temporal redundancy in the HDTV signals, relatively high levels of data compression can be achieved in most television signals.
One method by which spatial redundancy is removed from a television signal is to generate a discrete cosine transform representation of the image. This representation resolves an image frame into coefficient values representing its various spatial frequency components. Portions of an image which are the same from pixel to pixel or which exhibit a repetitive pattern are resolved from many pixel values to a relatively small number of frequency coefficient values. In addition, since the human eye is less sensitive to quantization errors in image components having high spatial frequencies than in image components having relatively low spatial frequencies, the high spatial frequency coefficients may be coarsely quantized to further reduce the amount of data used to represent the image.
Temporal redundancy in an image is removed by encoding only the areas of a given frame which are different from corresponding areas of a previously encoded frame. This is generally known as predictive coding. Temporal redundancy can be further exploited to achieve even higher levels of data compression by performing motion compensation. Using this scheme, before a image block is encoded, the blocks surrounding it in a previously encoded frame are searched for one block which most closely matches the block in the current frame. The current block is then subtracted from the matching block in the previous frame. The spatial frequency coefficient values are generated based upon the differences between the two image blocks. An exemplary video image compression system which employs motion-compensated predictive encoding techniques is that proposed by the Motion Picture Experts Group (MPEG) and described in the document entitled "Coded Representation of Picture and Audio Information" ISO-IEC/JTC1/SC2/WG11 N0010 MPEG 90/41 dated Jul. 25, 1990.
Other encoding techniques are also used in the MPEG system such as run-length coding, in which strings of identical values are encoded as a smaller number of values; and variable length coding, in which frequently occurring data values are assigned a digital code value having a fewer number of bits than less frequently occurring data values.
Whatever encoding technique is used the HDTV signal must be decoded before it can be displayed. For predictively encoded or motion compensated predictively encoded HDTV signals, the decoding apparatus may include one or more frame memories which hold images that have already been decoded. The pixel values held by these memories are used to reconstruct the predictively encoded data in the current frame.
Using MPEG encoding techniques, HDTV signals having a data rate of between 600 and 1200 Megabits per second (Mbps) can be compressed to produce a signal having a data rate of less than 20 Mbps. As with other terrestrial broadcast signals, consumers will want to be able to receive, display and record signals representing high resolution video images with little no loss of resolution.
On first analysis, the compression of HDTV signals should be beneficial for recording the signals on home use Video Tape Recorders (VTR's), since these units typically have only a limited bandwidth available for recording video signals. For example, a paper by C. Yamamitsu et al. entitled "A Study on Trick-plays For Digital VCR", IEEE Transactions on Consumer Electronics, Volume 37, No. 3, August, 1991, PP. 261-266, discloses a home use VCR having a recording rate of 27 Megabits per second (Mbps). A typical HDTV signal, prior to encoding, has a bit-rate of 600 Mbps. Compression methods such as MPEG can reduce these HDTV signals to have a bit-rate of approximately 18 Mbps without noticeably degrading the image quality when the compressed signal is expanded.
The problem with predictively encoded HDTV signals does not occur in recording or in normal playback modes but in trick-play modes such as fast-forward and reverse in which the video image is displayed at a higher rate than that at which it was recorded.
The problem is illustrated in FIGS. 1 and 2. FIG. 1 shows the track scanning sequence of pre-recorded video information during normal playback. As shown in FIG. 1, slices of the video image 110, 112, 114 and 116 are scanned in sequence as the tape head moves along the track 0. Each of these slices represents the same number of pixels in the reproduced image. As shown in the figured however, the amount of data in a slice may vary from slice-to-slice. This variation in the amount of data in a slice occurs because of the relative coding efficiencies of the slices in the original HDTV signal. Slices which represent still portions of a multi-frame image or which represent portions having relatively little variation may be encoded using a relatively small number of data values. Image portions containing a high level of detail and having no corresponding portions in previously encoded frames may require a significantly larger number of data values when they are encoded.
When the image signals are read from the tape during normal playback each slice of each frame is read from the tape in sequence. If the coding method which produced the HDTV signals used predictive coding techniques the recorded data includes both intra-frame coded data and predictively coded data. In normal playback as the data is removed from the tape the pixel values for the intra-frames are stored in a memory and are available for use in reconstructing the predicted frames.
FIG. 2 illustrates the track scanning sequence during a fast-forward trick-play mode. In this mode, instead of sequentially taking the slices 110, 112, 114 and 116 the slices recovered are in sequence 110, 118, 120, 122 and 124. As shown in this figure, some slices, for example, slices 112, 114, 116, 119 and 121, are never recovered from the tape. If some of these slices (e.g. 112 and 114) are from an intra-coded frame, then their corresponding data values will not be available when corresponding slices (e.g. 122 and 124) of a predicted frame, which is based on the intra-coded frame, are read from the tape. Consequently it may not be possible to properly reconstruct the predicted frame for display.
Because of the difficulties of recovering predicted frames in trick-play modes most of the proposed methods for recording digitally compressed (bit-rate reduced) signals have restricted the encoding techniques that are used to encode data for recording on VTR's to exclude predicted frames. Exemplary systems of this type are described in an article by C. Yamamitsu, et al. entitled "An Experimental Study for a Home-Use Digital VTR", IEEE Transactions on Consumer Electronics, Volume 35, No. 3, August 1989, PP. 450-457, and in a paper by J. Lee et al., entitled "A Study on New DCT-Based Bit Rate Reduction Algorithm and Variable Speed Playback for a Home-Use Digital VCR", IEEE Transactions on Consumer Electronics, Volume 38, No. 3, August 1992, PP. 236-242. As described above, since these systems do not use predicted frames, they cannot compress data with the efficiency of a system which does use predicted frames. Consequently, for the same compression ratios, they cannot achieve the same levels of detail in the reproduced image as a system, such as MPEG, which uses motion-compensated predictive coding techniques.